static int
unionfs_lock(void *v)
{
struct vop_lock_args *ap = v;
int error;
int flags;
struct vnode *lvp;
struct vnode *uvp;
struct unionfs_node *unp;
unp = VTOUNIONFS(ap->a_vp);
lvp = unp->un_lowervp;
uvp = unp->un_uppervp;
flags = ap->a_flags;
error = 0;
if (lvp != NULLVP) {
error = VOP_LOCK(lvp, flags);
}
if (error == 0 && uvp != NULLVP) {
error = VOP_LOCK(uvp, flags);
if (error != 0) {
VOP_UNLOCK(lvp);
}
}
return error;
}
/*
* Check that the vnode is still valid, and if so
* acquire requested lock.
*/
int
vn_lock(struct vnode *vp, int flags)
{
int error;
#if 0
KASSERT(vp->v_usecount > 0 || (vp->v_iflag & VI_ONWORKLST) != 0);
#endif
KASSERT((flags & ~(LK_SHARED|LK_EXCLUSIVE|LK_NOWAIT|LK_RETRY)) == 0);
KASSERT(!mutex_owned(vp->v_interlock));
#ifdef DIAGNOSTIC
if (wapbl_vphaswapbl(vp))
WAPBL_JUNLOCK_ASSERT(wapbl_vptomp(vp));
#endif
error = VOP_LOCK(vp, flags);
if ((flags & LK_RETRY) != 0 && error == ENOENT)
error = VOP_LOCK(vp, flags);
KASSERT((flags & LK_RETRY) == 0 || (flags & LK_NOWAIT) != 0 ||
error == 0);
return error;
}
int
nandfs_get_node_entry(struct nandfsmount *nmp, struct nandfs_inode **inode,
uint64_t ino, struct buf **bp)
{
struct nandfs_alloc_request req;
struct nandfs_mdt *mdt;
struct nandfs_node *ifile;
struct vnode *vp;
uint32_t index;
int error = 0;
req.entrynum = ino;
mdt = &nmp->nm_nandfsdev->nd_ifile_mdt;
ifile = nmp->nm_ifile_node;
vp = NTOV(ifile);
VOP_LOCK(vp, LK_EXCLUSIVE);
error = nandfs_get_entry_block(mdt, ifile, &req, &index, 0);
if (error) {
VOP_UNLOCK(vp, 0);
return (error);
}
*inode = ((struct nandfs_inode *) req.bp_entry->b_data) + index;
*bp = req.bp_entry;
VOP_UNLOCK(vp, 0);
return (0);
}
/* Generic write interface */
int
afs_osi_Write(struct osi_file *afile, afs_int32 offset, void *aptr,
afs_int32 asize)
{
unsigned int resid;
afs_int32 code;
AFS_STATCNT(osi_Write);
if (!afile)
osi_Panic("afs_osi_Write called with null afile");
if (offset != -1)
afile->offset = offset;
AFS_GUNLOCK();
VOP_LOCK(afile->vnode, LK_EXCLUSIVE | LK_RETRY);
code =
vn_rdwr(UIO_WRITE, afile->vnode, (caddr_t) aptr, asize, afile->offset,
AFS_UIOSYS, IO_UNIT, afs_osi_credp, &resid, osi_curproc());
VOP_UNLOCK(afile->vnode, 0);
AFS_GLOCK();
if (code == 0) {
code = asize - resid;
afile->offset += code;
if (afile->offset > afile->size)
afile->size = afile->offset;
} else
code = -1;
if (afile->proc)
(*afile->proc) (afile, code);
return code;
}
/*
* Purge the cache of dead entries
*
* This is extremely inefficient due to the fact that vgone() not only
* indirectly modifies the vnode cache, but may also sleep. We can
* neither hold pfs_vncache_mutex across a vgone() call, nor make any
* assumptions about the state of the cache after vgone() returns. In
* consequence, we must start over after every vgone() call, and keep
* trying until we manage to traverse the entire cache.
*
* The only way to improve this situation is to change the data structure
* used to implement the cache.
*/
static void
pfs_purge_locked(struct pfs_node *pn, bool force)
{
struct pfs_vdata *pvd;
struct vnode *vnp;
mtx_assert(&pfs_vncache_mutex, MA_OWNED);
pvd = pfs_vncache;
while (pvd != NULL) {
if (force || pvd->pvd_dead ||
(pn != NULL && pvd->pvd_pn == pn)) {
vnp = pvd->pvd_vnode;
vhold(vnp);
mtx_unlock(&pfs_vncache_mutex);
VOP_LOCK(vnp, LK_EXCLUSIVE);
vgone(vnp);
VOP_UNLOCK(vnp, 0);
mtx_lock(&pfs_vncache_mutex);
vdrop(vnp);
pvd = pfs_vncache;
} else {
pvd = pvd->pvd_next;
}
}
}
static int
_xfs_mkdir(
struct vop_mkdir_args /* {
struct vnode *a_dvp;
struct vnode **a_vpp;
struct componentname *a_cnp;
struct vattr *a_vap;
} */ *ap)
{
struct vnode *dvp = ap->a_dvp;
struct vattr *vap = ap->a_vap;
struct thread *td = curthread;
struct ucred *credp = td->td_ucred;
struct componentname *cnp = ap->a_cnp;
xfs_vnode_t *xvp;
xfs_vattr_t va;
int error;
memset(&va, 0, sizeof (va));
va.va_mask |= XFS_AT_MODE;
va.va_mode = vap->va_mode | S_IFDIR;
va.va_mask |= XFS_AT_TYPE;
xvp = NULL;
XVOP_MKDIR(VPTOXFSVP(dvp), cnp, &va, &xvp, credp, error);
if (error == 0) {
*ap->a_vpp = xvp->v_vnode;
VOP_LOCK(xvp->v_vnode, LK_EXCLUSIVE);
}
return (error);
}
int
nandfs_get_dat_bdescs(struct nandfs_device *nffsdev, struct nandfs_bdesc *bd,
uint32_t nmembs)
{
struct nandfs_node *dat_node;
uint64_t map;
uint32_t i;
int error = 0;
dat_node = nffsdev->nd_dat_node;
VOP_LOCK(NTOV(dat_node), LK_EXCLUSIVE);
for (i = 0; i < nmembs; i++) {
DPRINTF(CLEAN,
("%s: bd ino:%#jx oblk:%#jx blocknr:%#jx off:%#jx\n",
__func__, (uintmax_t)bd[i].bd_ino,
(uintmax_t)bd[i].bd_oblocknr, (uintmax_t)bd[i].bd_blocknr,
(uintmax_t)bd[i].bd_offset));
error = nandfs_bmap_lookup(dat_node, bd[i].bd_offset, &map);
if (error)
break;
bd[i].bd_blocknr = map;
}
VOP_UNLOCK(NTOV(dat_node), 0);
return (error);
}
int
nandfs_markgc_segment(struct nandfs_device *fsdev, uint64_t seg)
{
struct nandfs_node *su_node;
struct nandfs_segment_usage *su_usage;
struct buf *bp;
uint64_t blk, offset;
int error;
su_node = fsdev->nd_su_node;
VOP_LOCK(NTOV(su_node), LK_EXCLUSIVE);
nandfs_seg_usage_blk_offset(fsdev, seg, &blk, &offset);
error = nandfs_bread(su_node, blk, NOCRED, 0, &bp);
if (error) {
brelse(bp);
VOP_UNLOCK(NTOV(su_node), 0);
return (error);
}
su_usage = SU_USAGE_OFF(bp, offset);
MPASS((su_usage->su_flags & NANDFS_SEGMENT_USAGE_GC) == 0);
su_usage->su_flags |= NANDFS_SEGMENT_USAGE_GC;
brelse(bp);
VOP_UNLOCK(NTOV(su_node), 0);
DPRINTF(SEG, ("%s: seg:%#jx\n", __func__, (uintmax_t)seg));
return (0);
}
int
osi_UFSTruncate(struct osi_file *afile, afs_int32 asize)
{
struct vattr tvattr;
afs_int32 code;
struct osi_stat tstat;
AFS_STATCNT(osi_Truncate);
/*
* This routine only shrinks files, and most systems
* have very slow truncates, even when the file is already
* small enough. Check now and save some time.
*/
code = afs_osi_Stat(afile, &tstat);
if (code || tstat.size <= asize)
return code;
ObtainWriteLock(&afs_xosi, 321);
VATTR_NULL(&tvattr);
tvattr.va_size = asize;
AFS_GUNLOCK();
VOP_LOCK(afile->vnode, LK_EXCLUSIVE | LK_RETRY, curproc);
code = VOP_SETATTR(afile->vnode, &tvattr, afs_osi_credp, curproc);
VOP_UNLOCK(afile->vnode, 0, curproc);
AFS_GLOCK();
if (code == 0)
afile->size = asize;
ReleaseWriteLock(&afs_xosi);
return code;
}
static int
_xfs_symlink(
struct vop_symlink_args /* {
struct vnode *a_dvp;
struct vnode **a_vpp;
struct componentname *a_cnp;
struct vattr *a_vap;
char *a_target;
} */ *ap)
{
struct thread *td = curthread;
struct ucred *credp = td->td_ucred;
xfs_vnode_t *xvp;
xfs_vattr_t va;
int error;
memset(&va, 0, sizeof (va));
va.va_mask |= XFS_AT_MODE;
va.va_mode = ap->a_vap->va_mode | S_IFLNK;
va.va_mask |= XFS_AT_TYPE;
XVOP_SYMLINK(VPTOXFSVP(ap->a_dvp), ap->a_cnp, &va, ap->a_target,
&xvp, credp, error);
if (error == 0) {
*ap->a_vpp = xvp->v_vnode;
VOP_LOCK(xvp->v_vnode, LK_EXCLUSIVE);
}
return (error);
}
int
RUMP_VOP_LOCK(struct vnode *vp,
int flags)
{
int error;
rump_schedule();
error = VOP_LOCK(vp, flags);
rump_unschedule();
return error;
}
int
nandfs_node_create(struct nandfsmount *nmp, struct nandfs_node **node,
uint16_t mode)
{
struct nandfs_alloc_request req;
struct nandfs_device *nandfsdev;
struct nandfs_mdt *mdt;
struct nandfs_node *ifile;
struct nandfs_inode *inode;
struct vnode *vp;
uint32_t entry;
int error = 0;
nandfsdev = nmp->nm_nandfsdev;
mdt = &nandfsdev->nd_ifile_mdt;
ifile = nmp->nm_ifile_node;
vp = NTOV(ifile);
VOP_LOCK(vp, LK_EXCLUSIVE);
/* Allocate new inode in ifile */
req.entrynum = nandfsdev->nd_last_ino + 1;
error = nandfs_find_free_entry(mdt, ifile, &req);
if (error) {
VOP_UNLOCK(vp, 0);
return (error);
}
error = nandfs_get_entry_block(mdt, ifile, &req, &entry, 1);
if (error) {
VOP_UNLOCK(vp, 0);
return (error);
}
/* Inode initialization */
inode = ((struct nandfs_inode *) req.bp_entry->b_data) + entry;
nandfs_inode_init(inode, mode);
error = nandfs_alloc_entry(mdt, &req);
if (error) {
VOP_UNLOCK(vp, 0);
return (error);
}
VOP_UNLOCK(vp, 0);
nandfsdev->nd_last_ino = req.entrynum;
error = nandfs_get_node(nmp, req.entrynum, node);
DPRINTF(IFILE, ("%s: node: %p ino: %#jx\n",
__func__, node, (uintmax_t)((*node)->nn_ino)));
return (error);
}
static int
nandfs_process_bdesc(struct nandfs_device *nffsdev, struct nandfs_bdesc *bd,
uint64_t nmembs)
{
struct nandfs_node *dat_node;
struct buf *bp;
uint64_t i;
int error;
dat_node = nffsdev->nd_dat_node;
VOP_LOCK(NTOV(dat_node), LK_EXCLUSIVE);
for (i = 0; i < nmembs; i++) {
if (!bd[i].bd_alive)
continue;
DPRINTF(CLEAN, ("%s: idx %jx offset %jx\n",
__func__, i, bd[i].bd_offset));
if (bd[i].bd_level) {
error = nandfs_bread_meta(dat_node, bd[i].bd_offset,
NULL, 0, &bp);
if (error) {
nandfs_error("%s: cannot read dat node "
"level:%d\n", __func__, bd[i].bd_level);
brelse(bp);
VOP_UNLOCK(NTOV(dat_node), 0);
return (error);
}
nandfs_dirty_buf_meta(bp, 1);
nandfs_bmap_dirty_blocks(VTON(bp->b_vp), bp, 1);
} else {
error = nandfs_bread(dat_node, bd[i].bd_offset, NULL,
0, &bp);
if (error) {
nandfs_error("%s: cannot read dat node\n",
__func__);
brelse(bp);
VOP_UNLOCK(NTOV(dat_node), 0);
return (error);
}
nandfs_dirty_buf(bp, 1);
}
DPRINTF(CLEAN, ("%s: bp: %p\n", __func__, bp));
}
VOP_UNLOCK(NTOV(dat_node), 0);
return (0);
}
/*
* Insert the inode into the hash table, and return it locked.
*/
void
efs_ihashins(struct efs_inode *eip)
{
struct ihashhead *ipp;
KASSERT(mutex_owned(&efs_hashlock));
/* lock the inode, then put it on the appropriate hash list */
VOP_LOCK(EFS_ITOV(eip), LK_EXCLUSIVE);
mutex_enter(&efs_ihash_lock);
ipp = &ihashtbl[INOHASH(eip->ei_dev, eip->ei_number)];
LIST_INSERT_HEAD(ipp, eip, ei_hash);
mutex_exit(&efs_ihash_lock);
}
/*
* Insert the inode into the hash table, and return it locked.
*/
void
ulfs_ihashins(struct inode *ip)
{
struct ihashhead *ipp;
int error __diagused;
KASSERT(mutex_owned(&ulfs_hashlock));
/* lock the inode, then put it on the appropriate hash list */
error = VOP_LOCK(ITOV(ip), LK_EXCLUSIVE);
KASSERT(error == 0);
mutex_enter(&ulfs_ihash_lock);
ipp = &ihashtbl[INOHASH(ip->i_dev, ip->i_number)];
LIST_INSERT_HEAD(ipp, ip, i_hash);
mutex_exit(&ulfs_ihash_lock);
}
int
nandfs_node_destroy(struct nandfs_node *node)
{
struct nandfs_alloc_request req;
struct nandfsmount *nmp;
struct nandfs_mdt *mdt;
struct nandfs_node *ifile;
struct vnode *vp;
int error = 0;
nmp = node->nn_nmp;
req.entrynum = node->nn_ino;
mdt = &nmp->nm_nandfsdev->nd_ifile_mdt;
ifile = nmp->nm_ifile_node;
vp = NTOV(ifile);
DPRINTF(IFILE, ("%s: destroy node: %p ino: %#jx\n",
__func__, node, (uintmax_t)node->nn_ino));
VOP_LOCK(vp, LK_EXCLUSIVE);
error = nandfs_find_entry(mdt, ifile, &req);
if (error) {
nandfs_error("%s: finding entry error:%d node %p(%jx)",
__func__, error, node, node->nn_ino);
VOP_UNLOCK(vp, 0);
return (error);
}
nandfs_inode_destroy(&node->nn_inode);
error = nandfs_free_entry(mdt, &req);
if (error) {
nandfs_error("%s: freing entry error:%d node %p(%jx)",
__func__, error, node, node->nn_ino);
VOP_UNLOCK(vp, 0);
return (error);
}
VOP_UNLOCK(vp, 0);
DPRINTF(IFILE, ("%s: freed node %p ino %#jx\n",
__func__, node, (uintmax_t)node->nn_ino));
return (error);
}
/*
* Check that the vnode is still valid, and if so
* acquire requested lock.
*/
int
vn_lock(struct vnode *vp, int flags, struct proc *p)
{
int error;
if ((flags & LK_RECURSEFAIL) == 0)
flags |= LK_CANRECURSE;
do {
if (vp->v_flag & VXLOCK) {
vp->v_flag |= VXWANT;
tsleep(vp, PINOD, "vn_lock", 0);
error = ENOENT;
} else {
error = VOP_LOCK(vp, flags, p);
if (error == 0)
return (error);
}
} while (flags & LK_RETRY);
return (error);
}
int
nandfs_get_dat_vinfo(struct nandfs_device *nandfsdev, struct nandfs_vinfo *vinfo,
uint32_t nmembs)
{
struct nandfs_node *dat;
struct nandfs_mdt *mdt;
struct nandfs_alloc_request req;
struct nandfs_dat_entry *dat_entry;
uint32_t i, idx;
int error = 0;
dat = nandfsdev->nd_dat_node;
mdt = &nandfsdev->nd_dat_mdt;
DPRINTF(DAT, ("%s: nmembs %#x\n", __func__, nmembs));
VOP_LOCK(NTOV(dat), LK_EXCLUSIVE);
for (i = 0; i < nmembs; i++) {
req.entrynum = vinfo[i].nvi_vblocknr;
error = nandfs_get_entry_block(mdt, dat,&req, &idx, 0);
if (error)
break;
dat_entry = ((struct nandfs_dat_entry *) req.bp_entry->b_data);
vinfo[i].nvi_start = dat_entry[idx].de_start;
vinfo[i].nvi_end = dat_entry[idx].de_end;
vinfo[i].nvi_blocknr = dat_entry[idx].de_blocknr;
DPRINTF(DAT, ("%s: vinfo: %jx[%jx-%jx]->%jx\n",
__func__, vinfo[i].nvi_vblocknr, vinfo[i].nvi_start,
vinfo[i].nvi_end, vinfo[i].nvi_blocknr));
brelse(req.bp_entry);
}
VOP_UNLOCK(NTOV(dat), 0);
return (error);
}
/*
* Check that the vnode is still valid, and if so
* acquire requested lock.
*/
int
vn_lock(struct vnode *vp, int flags)
{
int error;
#if 0
KASSERT(vp->v_usecount > 0 || (vp->v_iflag & VI_ONWORKLST) != 0);
#endif
KASSERT((flags & ~(LK_SHARED|LK_EXCLUSIVE|LK_NOWAIT|LK_RETRY)) == 0);
KASSERT(!mutex_owned(vp->v_interlock));
#ifdef DIAGNOSTIC
if (wapbl_vphaswapbl(vp))
WAPBL_JUNLOCK_ASSERT(wapbl_vptomp(vp));
#endif
do {
/*
* XXX PR 37706 forced unmount of file systems is unsafe.
* Race between vclean() and this the remaining problem.
*/
mutex_enter(vp->v_interlock);
if (vp->v_iflag & VI_XLOCK) {
if (flags & LK_NOWAIT) {
mutex_exit(vp->v_interlock);
return EBUSY;
}
vwait(vp, VI_XLOCK);
mutex_exit(vp->v_interlock);
error = ENOENT;
} else {
mutex_exit(vp->v_interlock);
error = VOP_LOCK(vp, (flags & ~LK_RETRY));
if (error == 0 || error == EDEADLK || error == EBUSY)
return (error);
}
} while (flags & LK_RETRY);
return (error);
}
int
nandfs_vblock_end(struct nandfs_device *nandfsdev, nandfs_daddr_t vblock)
{
struct nandfs_node *dat;
struct nandfs_mdt *mdt;
struct nandfs_alloc_request req;
struct nandfs_dat_entry *dat_entry;
uint64_t end;
uint32_t entry;
int locked, error;
dat = nandfsdev->nd_dat_node;
mdt = &nandfsdev->nd_dat_mdt;
end = nandfsdev->nd_last_cno;
locked = NANDFS_VOP_ISLOCKED(NTOV(dat));
if (!locked)
VOP_LOCK(NTOV(dat), LK_EXCLUSIVE);
req.entrynum = vblock;
error = nandfs_get_entry_block(mdt, dat, &req, &entry, 0);
if (!error) {
dat_entry = (struct nandfs_dat_entry *) req.bp_entry->b_data;
dat_entry[entry].de_end = end;
DPRINTF(DAT, ("%s: end vblock %#jx at checkpoint %#jx\n",
__func__, (uintmax_t)vblock, (uintmax_t)end));
/*
* It is mostly called from syncer() so
* we want to force making buf dirty
*/
error = nandfs_dirty_buf(req.bp_entry, 1);
}
if (!locked)
VOP_UNLOCK(NTOV(dat), 0);
return (error);
}
int
nandfs_vblock_free(struct nandfs_device *nandfsdev, nandfs_daddr_t vblock)
{
struct nandfs_node *dat;
struct nandfs_mdt *mdt;
struct nandfs_alloc_request req;
int error;
dat = nandfsdev->nd_dat_node;
mdt = &nandfsdev->nd_dat_mdt;
VOP_LOCK(NTOV(dat), LK_EXCLUSIVE);
req.entrynum = vblock;
error = nandfs_find_entry(mdt, dat, &req);
if (!error) {
DPRINTF(DAT, ("%s: vblk %#jx\n", __func__, (uintmax_t)vblock));
nandfs_free_entry(mdt, &req);
}
VOP_UNLOCK(NTOV(dat), 0);
return (error);
}
int
nandfs_get_seg_stat(struct nandfs_device *nandfsdev,
struct nandfs_seg_stat *nss)
{
struct nandfs_sufile_header *suhdr;
struct nandfs_node *su_node;
struct buf *bp;
int err;
su_node = nandfsdev->nd_su_node;
NANDFS_WRITELOCK(nandfsdev);
VOP_LOCK(NTOV(su_node), LK_SHARED);
err = nandfs_bread(nandfsdev->nd_su_node, 0, NOCRED, 0, &bp);
if (err) {
brelse(bp);
VOP_UNLOCK(NTOV(su_node), 0);
NANDFS_WRITEUNLOCK(nandfsdev);
return (-1);
}
suhdr = (struct nandfs_sufile_header *)bp->b_data;
nss->nss_nsegs = nandfsdev->nd_fsdata.f_nsegments;
nss->nss_ncleansegs = suhdr->sh_ncleansegs;
nss->nss_ndirtysegs = suhdr->sh_ndirtysegs;
nss->nss_ctime = 0;
nss->nss_nongc_ctime = nandfsdev->nd_ts.tv_sec;
nss->nss_prot_seq = nandfsdev->nd_seg_sequence;
brelse(bp);
VOP_UNLOCK(NTOV(su_node), 0);
NANDFS_WRITEUNLOCK(nandfsdev);
return (0);
}
/*
* We need to process our own vnode lock and then clear the
* interlock flag as it applies only to our vnode, not the
* vnodes below us on the stack.
*/
static int
null_lock(struct vop_lock1_args *ap)
{
struct vnode *vp = ap->a_vp;
int flags = ap->a_flags;
struct null_node *nn;
struct vnode *lvp;
int error;
if ((flags & LK_INTERLOCK) == 0) {
VI_LOCK(vp);
ap->a_flags = flags |= LK_INTERLOCK;
}
nn = VTONULL(vp);
/*
* If we're still active we must ask the lower layer to
* lock as ffs has special lock considerations in it's
* vop lock.
*/
if (nn != NULL && (lvp = NULLVPTOLOWERVP(vp)) != NULL) {
VI_LOCK_FLAGS(lvp, MTX_DUPOK);
VI_UNLOCK(vp);
/*
* We have to hold the vnode here to solve a potential
* reclaim race. If we're forcibly vgone'd while we
* still have refs, a thread could be sleeping inside
* the lowervp's vop_lock routine. When we vgone we will
* drop our last ref to the lowervp, which would allow it
* to be reclaimed. The lowervp could then be recycled,
* in which case it is not legal to be sleeping in it's VOP.
* We prevent it from being recycled by holding the vnode
* here.
*/
vholdl(lvp);
error = VOP_LOCK(lvp, flags);
/*
* We might have slept to get the lock and someone might have
* clean our vnode already, switching vnode lock from one in
* lowervp to v_lock in our own vnode structure. Handle this
* case by reacquiring correct lock in requested mode.
*/
if (VTONULL(vp) == NULL && error == 0) {
ap->a_flags &= ~(LK_TYPE_MASK | LK_INTERLOCK);
switch (flags & LK_TYPE_MASK) {
case LK_SHARED:
ap->a_flags |= LK_SHARED;
break;
case LK_UPGRADE:
case LK_EXCLUSIVE:
ap->a_flags |= LK_EXCLUSIVE;
break;
default:
panic("Unsupported lock request %d\n",
ap->a_flags);
}
VOP_UNLOCK(lvp, 0);
error = vop_stdlock(ap);
}
vdrop(lvp);
} else
error = vop_stdlock(ap);
return (error);
}
int
afs_MemRead(struct vcache *avc, struct uio *auio,
afs_ucred_t *acred, daddr_t albn, struct buf **abpp,
int noLock)
{
afs_size_t totalLength;
afs_size_t transferLength;
afs_size_t filePos;
afs_size_t offset, tlen;
afs_size_t len = 0;
afs_int32 trimlen;
struct dcache *tdc = 0;
afs_int32 error, trybusy = 1;
afs_int32 code;
struct vrequest *treq = NULL;
#ifdef AFS_DARWIN80_ENV
uio_t tuiop = NULL;
#else
struct uio tuio;
struct uio *tuiop = &tuio;
struct iovec *tvec;
memset(&tuio, 0, sizeof(tuio));
#endif
AFS_STATCNT(afs_MemRead);
if (avc->vc_error)
return EIO;
/* check that we have the latest status info in the vnode cache */
if ((code = afs_CreateReq(&treq, acred)))
return code;
if (!noLock) {
code = afs_VerifyVCache(avc, treq);
if (code) {
code = afs_CheckCode(code, treq, 8); /* failed to get it */
afs_DestroyReq(treq);
return code;
}
}
#ifndef AFS_VM_RDWR_ENV
if (AFS_NFSXLATORREQ(acred)) {
if (!afs_AccessOK
(avc, PRSFS_READ, treq,
CHECK_MODE_BITS | CMB_ALLOW_EXEC_AS_READ)) {
code = afs_CheckCode(EACCES, treq, 9);
afs_DestroyReq(treq);
return code;
}
}
#endif
#ifndef AFS_DARWIN80_ENV
tvec = (struct iovec *)osi_AllocSmallSpace(sizeof(struct iovec));
memset(tvec, 0, sizeof(struct iovec));
#endif
totalLength = AFS_UIO_RESID(auio);
filePos = AFS_UIO_OFFSET(auio);
afs_Trace4(afs_iclSetp, CM_TRACE_READ, ICL_TYPE_POINTER, avc,
ICL_TYPE_OFFSET, ICL_HANDLE_OFFSET(filePos), ICL_TYPE_INT32,
totalLength, ICL_TYPE_OFFSET,
ICL_HANDLE_OFFSET(avc->f.m.Length));
error = 0;
transferLength = 0;
if (!noLock)
ObtainReadLock(&avc->lock);
#if defined(AFS_TEXT_ENV) && !defined(AFS_VM_RDWR_ENV)
if (avc->flushDV.high == AFS_MAXDV && avc->flushDV.low == AFS_MAXDV) {
hset(avc->flushDV, avc->f.m.DataVersion);
}
#endif
/*
* Locks held:
* avc->lock(R)
*/
/* This bit is bogus. We're checking to see if the read goes past the
* end of the file. If so, we should be zeroing out all of the buffers
* that the client has passed into us (there is a danger that we may leak
* kernel memory if we do not). However, this behaviour is disabled by
* not setting len before this segment runs, and by setting len to 0
* immediately we enter it. In addition, we also need to check for a read
* which partially goes off the end of the file in the while loop below.
*/
if (filePos >= avc->f.m.Length) {
if (len > AFS_ZEROS)
len = sizeof(afs_zeros); /* and in 0 buffer */
len = 0;
#ifdef AFS_DARWIN80_ENV
trimlen = len;
tuiop = afsio_darwin_partialcopy(auio, trimlen);
#else
afsio_copy(auio, &tuio, tvec);
trimlen = len;
afsio_trim(&tuio, trimlen);
#endif
AFS_UIOMOVE(afs_zeros, trimlen, UIO_READ, tuiop, code);
}
while (avc->f.m.Length > 0 && totalLength > 0) {
/* read all of the cached info */
if (filePos >= avc->f.m.Length)
break; /* all done */
if (noLock) {
if (tdc) {
ReleaseReadLock(&tdc->lock);
afs_PutDCache(tdc);
}
tdc = afs_FindDCache(avc, filePos);
if (tdc) {
ObtainReadLock(&tdc->lock);
offset = filePos - AFS_CHUNKTOBASE(tdc->f.chunk);
len = tdc->f.chunkBytes - offset;
}
} else {
int versionOk;
/* a tricky question: does the presence of the DFFetching flag
* mean that we're fetching the latest version of the file? No.
* The server could update the file as soon as the fetch responsible
* for the setting of the DFFetching flag completes.
*
* However, the presence of the DFFetching flag (visible under
* a dcache read lock since it is set and cleared only under a
* dcache write lock) means that we're fetching as good a version
* as was known to this client at the time of the last call to
* afs_VerifyVCache, since the latter updates the stat cache's
* m.DataVersion field under a vcache write lock, and from the
* time that the DFFetching flag goes on in afs_GetDCache (before
* the fetch starts), to the time it goes off (after the fetch
* completes), afs_GetDCache keeps at least a read lock on the
* vcache entry.
*
* This means that if the DFFetching flag is set, we can use that
* data for any reads that must come from the current version of
* the file (current == m.DataVersion).
*
* Another way of looking at this same point is this: if we're
* fetching some data and then try do an afs_VerifyVCache, the
* VerifyVCache operation will not complete until after the
* DFFetching flag is turned off and the dcache entry's f.versionNo
* field is updated.
*
* Note, by the way, that if DFFetching is set,
* m.DataVersion > f.versionNo (the latter is not updated until
* after the fetch completes).
*/
if (tdc) {
ReleaseReadLock(&tdc->lock);
afs_PutDCache(tdc); /* before reusing tdc */
}
#ifdef STRUCT_TASK_STRUCT_HAS_CRED
try_background:
#endif
tdc = afs_GetDCache(avc, filePos, treq, &offset, &len, 2);
ObtainReadLock(&tdc->lock);
/* now, first try to start transfer, if we'll need the data. If
* data already coming, we don't need to do this, obviously. Type
* 2 requests never return a null dcache entry, btw.
*/
if (!(tdc->dflags & DFFetching)
&& !hsame(avc->f.m.DataVersion, tdc->f.versionNo)) {
/* have cache entry, it is not coming in now,
* and we'll need new data */
tagain:
#ifdef STRUCT_TASK_STRUCT_HAS_CRED
if (trybusy && (!afs_BBusy() || (afs_protocols & VICEP_ACCESS))) {
#else
if (trybusy && !afs_BBusy()) {
#endif
struct brequest *bp;
/* daemon is not busy */
ObtainSharedLock(&tdc->mflock, 665);
if (!(tdc->mflags & DFFetchReq)) {
int dontwait = B_DONTWAIT;
/* start the daemon (may already be running, however) */
UpgradeSToWLock(&tdc->mflock, 666);
tdc->mflags |= DFFetchReq;
#ifdef STRUCT_TASK_STRUCT_HAS_CRED
if (afs_protocols & VICEP_ACCESS)
dontwait = 0;
#endif
bp = afs_BQueue(BOP_FETCH, avc, dontwait, 0, acred,
(afs_size_t) filePos, (afs_size_t) 0,
tdc, (void *)0, (void *)0);
if (!bp) {
tdc->mflags &= ~DFFetchReq;
trybusy = 0; /* Avoid bkg daemon since they're too busy */
ReleaseWriteLock(&tdc->mflock);
goto tagain;
}
ConvertWToSLock(&tdc->mflock);
/* don't use bp pointer! */
}
code = 0;
ConvertSToRLock(&tdc->mflock);
while (!code && tdc->mflags & DFFetchReq) {
afs_Trace4(afs_iclSetp, CM_TRACE_DCACHEWAIT,
ICL_TYPE_STRING, __FILE__, ICL_TYPE_INT32,
__LINE__, ICL_TYPE_POINTER, tdc,
ICL_TYPE_INT32, tdc->dflags);
/* don't need waiting flag on this one */
ReleaseReadLock(&tdc->mflock);
ReleaseReadLock(&tdc->lock);
ReleaseReadLock(&avc->lock);
code = afs_osi_SleepSig(&tdc->validPos);
ObtainReadLock(&avc->lock);
ObtainReadLock(&tdc->lock);
ObtainReadLock(&tdc->mflock);
}
ReleaseReadLock(&tdc->mflock);
if (code) {
error = code;
break;
}
}
}
/* now data may have started flowing in (if DFFetching is on). If
* data is now streaming in, then wait for some interesting stuff.
*/
code = 0;
while (!code && (tdc->dflags & DFFetching)
&& tdc->validPos <= filePos) {
/* too early: wait for DFFetching flag to vanish,
* or data to appear */
afs_Trace4(afs_iclSetp, CM_TRACE_DCACHEWAIT, ICL_TYPE_STRING,
__FILE__, ICL_TYPE_INT32, __LINE__,
ICL_TYPE_POINTER, tdc, ICL_TYPE_INT32,
tdc->dflags);
ReleaseReadLock(&tdc->lock);
ReleaseReadLock(&avc->lock);
code = afs_osi_SleepSig(&tdc->validPos);
ObtainReadLock(&avc->lock);
ObtainReadLock(&tdc->lock);
}
if (code) {
error = code;
break;
}
/* fetching flag gone, data is here, or we never tried
* (BBusy for instance) */
len = tdc->validPos - filePos;
versionOk = hsame(avc->f.m.DataVersion, tdc->f.versionNo) ? 1 : 0;
if (tdc->dflags & DFFetching) {
/* still fetching, some new data is here:
* compute length and offset */
offset = filePos - AFS_CHUNKTOBASE(tdc->f.chunk);
} else {
/* no longer fetching, verify data version
* (avoid new GetDCache call) */
if (versionOk && len > 0) {
offset = filePos - AFS_CHUNKTOBASE(tdc->f.chunk);
} else {
/* don't have current data, so get it below */
afs_Trace3(afs_iclSetp, CM_TRACE_VERSIONNO,
ICL_TYPE_INT64, ICL_HANDLE_OFFSET(filePos),
ICL_TYPE_HYPER, &avc->f.m.DataVersion,
ICL_TYPE_HYPER, &tdc->f.versionNo);
#if 0
#ifdef STRUCT_TASK_STRUCT_HAS_CRED
if (afs_protocols & VICEP_ACCESS) {
printf("afs_read: DV mismatch? %d instead of %d for %u.%u.%u\n", tdc->f.versionNo.low, avc->f.m.DataVersion.low, avc->f.fid.Fid.Volume, avc->f.fid.Fid.Vnode, avc->f.fid.Fid.Unique);
printf("afs_read: validPos %llu filePos %llu totalLength %lld m.Length %llu noLock %d\n", tdc->validPos, filePos, totalLength, avc->f.m.Length, noLock);
printf("afs_read: or len too low? %lld for %u.%u.%u\n", len, avc->f.fid.Fid.Volume, avc->f.fid.Fid.Vnode, avc->f.fid.Fid.Unique);
}
#endif
#endif
ReleaseReadLock(&tdc->lock);
afs_PutDCache(tdc);
tdc = NULL;
}
}
if (!tdc) {
#ifdef STRUCT_TASK_STRUCT_HAS_CRED
if (afs_protocols & VICEP_ACCESS) { /* avoid foreground fetch */
if (!versionOk) {
printf("afs_read: avoid forground %u.%u.%u\n", avc->f.fid.Fid.Volume, avc->f.fid.Fid.Vnode, avc->f.fid.Fid.Unique);
goto try_background;
}
#if 0
printf("afs_read: forground %u.%u.%u\n", avc->f.fid.Fid.Volume, avc->f.fid.Fid.Vnode, avc->f.fid.Fid.Unique);
#endif
}
#endif
/* If we get here, it was not possible to start the
* background daemon. With flag == 1 afs_GetDCache
* does the FetchData rpc synchronously.
*/
ReleaseReadLock(&avc->lock);
tdc = afs_GetDCache(avc, filePos, treq, &offset, &len, 1);
ObtainReadLock(&avc->lock);
if (tdc)
ObtainReadLock(&tdc->lock);
}
}
afs_Trace3(afs_iclSetp, CM_TRACE_VNODEREAD, ICL_TYPE_POINTER, tdc,
ICL_TYPE_OFFSET, ICL_HANDLE_OFFSET(offset),
ICL_TYPE_OFFSET, ICL_HANDLE_OFFSET(len));
if (!tdc) {
error = EIO;
break;
}
/*
* Locks held:
* avc->lock(R)
* tdc->lock(R)
*/
if (len > totalLength)
len = totalLength; /* will read len bytes */
if (len <= 0) { /* shouldn't get here if DFFetching is on */
/* read past the end of a chunk, may not be at next chunk yet, and yet
* also not at eof, so may have to supply fake zeros */
len = AFS_CHUNKTOSIZE(tdc->f.chunk) - offset; /* bytes left in chunk addr space */
if (len > totalLength)
len = totalLength; /* and still within xfr request */
tlen = avc->f.m.Length - offset; /* and still within file */
if (len > tlen)
len = tlen;
if (len > AFS_ZEROS)
len = sizeof(afs_zeros); /* and in 0 buffer */
#ifdef AFS_DARWIN80_ENV
trimlen = len;
tuiop = afsio_darwin_partialcopy(auio, trimlen);
#else
afsio_copy(auio, &tuio, tvec);
trimlen = len;
afsio_trim(&tuio, trimlen);
#endif
AFS_UIOMOVE(afs_zeros, trimlen, UIO_READ, tuiop, code);
if (code) {
error = code;
break;
}
} else {
/* get the data from the mem cache */
/* mung uio structure to be right for this transfer */
#ifdef AFS_DARWIN80_ENV
trimlen = len;
tuiop = afsio_darwin_partialcopy(auio, trimlen);
uio_setoffset(tuiop, offset);
#else
afsio_copy(auio, &tuio, tvec);
trimlen = len;
afsio_trim(&tuio, trimlen);
tuio.afsio_offset = offset;
#endif
code = afs_MemReadUIO(&tdc->f.inode, tuiop);
if (code) {
error = code;
break;
}
}
/* otherwise we've read some, fixup length, etc and continue with next seg */
len = len - AFS_UIO_RESID(tuiop); /* compute amount really transferred */
trimlen = len;
afsio_skip(auio, trimlen); /* update input uio structure */
totalLength -= len;
transferLength += len;
filePos += len;
if (len <= 0)
break; /* surprise eof */
#ifdef AFS_DARWIN80_ENV
if (tuiop) {
uio_free(tuiop);
tuiop = 0;
}
#endif
} /* the whole while loop */
/*
* Locks held:
* avc->lock(R)
* tdc->lock(R) if tdc
*/
/* if we make it here with tdc non-zero, then it is the last chunk we
* dealt with, and we have to release it when we're done. We hold on
* to it in case we need to do a prefetch.
*/
if (tdc) {
ReleaseReadLock(&tdc->lock);
/*
* try to queue prefetch, if needed. If DataVersion is zero there
* should not be any more: files with DV 0 never have been stored
* on the fileserver, symbolic links and directories never require
* more than a single chunk.
*/
if (!noLock && !(hiszero(avc->f.m.DataVersion)) &&
#ifndef AFS_VM_RDWR_ENV
afs_preCache
#else
1
#endif
) {
afs_PrefetchChunk(avc, tdc, acred, treq);
}
afs_PutDCache(tdc);
}
if (!noLock)
ReleaseReadLock(&avc->lock);
#ifdef AFS_DARWIN80_ENV
if (tuiop)
uio_free(tuiop);
#else
osi_FreeSmallSpace(tvec);
#endif
error = afs_CheckCode(error, treq, 10);
afs_DestroyReq(treq);
return error;
}
/* called with the dcache entry triggering the fetch, the vcache entry involved,
* and a vrequest for the read call. Marks the dcache entry as having already
* triggered a prefetch, starts the prefetch going and sets the DFFetchReq
* flag in the prefetched block, so that the next call to read knows to wait
* for the daemon to start doing things.
*
* This function must be called with the vnode at least read-locked, and
* no locks on the dcache, because it plays around with dcache entries.
*/
void
afs_PrefetchChunk(struct vcache *avc, struct dcache *adc,
afs_ucred_t *acred, struct vrequest *areq)
{
struct dcache *tdc;
afs_size_t offset;
afs_size_t j1, j2; /* junk vbls for GetDCache to trash */
offset = adc->f.chunk + 1; /* next chunk we'll need */
offset = AFS_CHUNKTOBASE(offset); /* base of next chunk */
ObtainReadLock(&adc->lock);
ObtainSharedLock(&adc->mflock, 662);
if (offset < avc->f.m.Length && !(adc->mflags & DFNextStarted)
&& !afs_BBusy()) {
struct brequest *bp;
UpgradeSToWLock(&adc->mflock, 663);
adc->mflags |= DFNextStarted; /* we've tried to prefetch for this guy */
ReleaseWriteLock(&adc->mflock);
ReleaseReadLock(&adc->lock);
tdc = afs_GetDCache(avc, offset, areq, &j1, &j2, 2); /* type 2 never returns 0 */
/*
* In disconnected mode, type 2 can return 0 because it doesn't
* make any sense to allocate a dcache we can never fill
*/
if (tdc == NULL)
return;
ObtainSharedLock(&tdc->mflock, 651);
if (!(tdc->mflags & DFFetchReq)) {
/* ask the daemon to do the work */
UpgradeSToWLock(&tdc->mflock, 652);
tdc->mflags |= DFFetchReq; /* guaranteed to be cleared by BKG or GetDCache */
/* last parm (1) tells bkg daemon to do an afs_PutDCache when it is done,
* since we don't want to wait for it to finish before doing so ourselves.
*/
bp = afs_BQueue(BOP_FETCH, avc, B_DONTWAIT, 0, acred,
(afs_size_t) offset, (afs_size_t) 1, tdc,
(void *)0, (void *)0);
if (!bp) {
/* Bkg table full; just abort non-important prefetching to avoid deadlocks */
tdc->mflags &= ~DFFetchReq;
ReleaseWriteLock(&tdc->mflock);
afs_PutDCache(tdc);
/*
* DCLOCKXXX: This is a little sketchy, since someone else
* could have already started a prefetch.. In practice,
* this probably doesn't matter; at most it would cause an
* extra slot in the BKG table to be used up when someone
* prefetches this for the second time.
*/
ObtainReadLock(&adc->lock);
ObtainWriteLock(&adc->mflock, 664);
adc->mflags &= ~DFNextStarted;
ReleaseWriteLock(&adc->mflock);
ReleaseReadLock(&adc->lock);
} else {
ReleaseWriteLock(&tdc->mflock);
}
} else {
ReleaseSharedLock(&tdc->mflock);
afs_PutDCache(tdc);
}
} else {
ReleaseSharedLock(&adc->mflock);
ReleaseReadLock(&adc->lock);
}
}
int
afs_UFSRead(struct vcache *avc, struct uio *auio,
afs_ucred_t *acred, daddr_t albn, struct buf **abpp,
int noLock)
{
afs_size_t totalLength;
afs_size_t transferLength;
afs_size_t filePos;
afs_size_t offset, tlen;
afs_size_t len = 0;
afs_int32 trimlen;
struct dcache *tdc = 0;
afs_int32 error;
struct osi_file *tfile;
afs_int32 code;
int trybusy = 1;
struct vrequest *treq = NULL;
#ifdef AFS_DARWIN80_ENV
uio_t tuiop=NULL;
#else
struct uio tuio;
struct uio *tuiop = &tuio;
struct iovec *tvec;
memset(&tuio, 0, sizeof(tuio));
#endif
AFS_STATCNT(afs_UFSRead);
if (avc && avc->vc_error)
return EIO;
AFS_DISCON_LOCK();
/* check that we have the latest status info in the vnode cache */
if ((code = afs_CreateReq(&treq, acred)))
return code;
if (!noLock) {
if (!avc)
osi_Panic("null avc in afs_UFSRead");
else {
code = afs_VerifyVCache(avc, treq);
if (code) {
code = afs_CheckCode(code, treq, 11); /* failed to get it */
afs_DestroyReq(treq);
AFS_DISCON_UNLOCK();
return code;
}
}
}
#ifndef AFS_VM_RDWR_ENV
if (AFS_NFSXLATORREQ(acred)) {
if (!afs_AccessOK
(avc, PRSFS_READ, treq,
CHECK_MODE_BITS | CMB_ALLOW_EXEC_AS_READ)) {
AFS_DISCON_UNLOCK();
code = afs_CheckCode(EACCES, treq, 12);
afs_DestroyReq(treq);
return code;
}
}
#endif
#ifndef AFS_DARWIN80_ENV
tvec = (struct iovec *)osi_AllocSmallSpace(sizeof(struct iovec));
memset(tvec, 0, sizeof(struct iovec));
#endif
totalLength = AFS_UIO_RESID(auio);
filePos = AFS_UIO_OFFSET(auio);
afs_Trace4(afs_iclSetp, CM_TRACE_READ, ICL_TYPE_POINTER, avc,
ICL_TYPE_OFFSET, ICL_HANDLE_OFFSET(filePos), ICL_TYPE_INT32,
totalLength, ICL_TYPE_OFFSET,
ICL_HANDLE_OFFSET(avc->f.m.Length));
error = 0;
transferLength = 0;
if (!noLock)
ObtainReadLock(&avc->lock);
#if defined(AFS_TEXT_ENV) && !defined(AFS_VM_RDWR_ENV)
if (avc->flushDV.high == AFS_MAXDV && avc->flushDV.low == AFS_MAXDV) {
hset(avc->flushDV, avc->f.m.DataVersion);
}
#endif
/* This bit is bogus. We're checking to see if the read goes past the
* end of the file. If so, we should be zeroing out all of the buffers
* that the client has passed into us (there is a danger that we may leak
* kernel memory if we do not). However, this behaviour is disabled by
* not setting len before this segment runs, and by setting len to 0
* immediately we enter it. In addition, we also need to check for a read
* which partially goes off the end of the file in the while loop below.
*/
if (filePos >= avc->f.m.Length) {
if (len > AFS_ZEROS)
len = sizeof(afs_zeros); /* and in 0 buffer */
len = 0;
#ifdef AFS_DARWIN80_ENV
trimlen = len;
tuiop = afsio_darwin_partialcopy(auio, trimlen);
#else
afsio_copy(auio, &tuio, tvec);
trimlen = len;
afsio_trim(&tuio, trimlen);
#endif
AFS_UIOMOVE(afs_zeros, trimlen, UIO_READ, tuiop, code);
}
while (avc->f.m.Length > 0 && totalLength > 0) {
/* read all of the cached info */
if (filePos >= avc->f.m.Length)
break; /* all done */
if (noLock) {
if (tdc) {
ReleaseReadLock(&tdc->lock);
afs_PutDCache(tdc);
}
tdc = afs_FindDCache(avc, filePos);
if (tdc) {
ObtainReadLock(&tdc->lock);
offset = filePos - AFS_CHUNKTOBASE(tdc->f.chunk);
len = tdc->validPos - filePos;
}
} else {
int versionOk;
/* a tricky question: does the presence of the DFFetching flag
* mean that we're fetching the latest version of the file? No.
* The server could update the file as soon as the fetch responsible
* for the setting of the DFFetching flag completes.
*
* However, the presence of the DFFetching flag (visible under
* a dcache read lock since it is set and cleared only under a
* dcache write lock) means that we're fetching as good a version
* as was known to this client at the time of the last call to
* afs_VerifyVCache, since the latter updates the stat cache's
* m.DataVersion field under a vcache write lock, and from the
* time that the DFFetching flag goes on in afs_GetDCache (before
* the fetch starts), to the time it goes off (after the fetch
* completes), afs_GetDCache keeps at least a read lock on the
* vcache entry.
*
* This means that if the DFFetching flag is set, we can use that
* data for any reads that must come from the current version of
* the file (current == m.DataVersion).
*
* Another way of looking at this same point is this: if we're
* fetching some data and then try do an afs_VerifyVCache, the
* VerifyVCache operation will not complete until after the
* DFFetching flag is turned off and the dcache entry's f.versionNo
* field is updated.
*
* Note, by the way, that if DFFetching is set,
* m.DataVersion > f.versionNo (the latter is not updated until
* after the fetch completes).
*/
if (tdc) {
ReleaseReadLock(&tdc->lock);
afs_PutDCache(tdc); /* before reusing tdc */
}
#ifdef STRUCT_TASK_STRUCT_HAS_CRED
try_background:
#endif
tdc = afs_GetDCache(avc, filePos, treq, &offset, &len, 2);
if (!tdc) {
error = ENETDOWN;
break;
}
ObtainReadLock(&tdc->lock);
/* now, first try to start transfer, if we'll need the data. If
* data already coming, we don't need to do this, obviously. Type
* 2 requests never return a null dcache entry, btw. */
if (!(tdc->dflags & DFFetching)
&& !hsame(avc->f.m.DataVersion, tdc->f.versionNo)) {
/* have cache entry, it is not coming in now, and we'll need new data */
tagain:
#ifdef STRUCT_TASK_STRUCT_HAS_CRED
if (trybusy && (!afs_BBusy() || (afs_protocols & VICEP_ACCESS))) {
#else
if (trybusy && !afs_BBusy()) {
#endif
struct brequest *bp;
/* daemon is not busy */
ObtainSharedLock(&tdc->mflock, 667);
if (!(tdc->mflags & DFFetchReq)) {
int dontwait = B_DONTWAIT;
UpgradeSToWLock(&tdc->mflock, 668);
tdc->mflags |= DFFetchReq;
#ifdef STRUCT_TASK_STRUCT_HAS_CRED
if (afs_protocols & VICEP_ACCESS)
dontwait = 0;
#endif
bp = afs_BQueue(BOP_FETCH, avc, dontwait, 0, acred,
(afs_size_t) filePos, (afs_size_t) 0,
tdc, (void *)0, (void *)0);
if (!bp) {
/* Bkg table full; retry deadlocks */
tdc->mflags &= ~DFFetchReq;
trybusy = 0; /* Avoid bkg daemon since they're too busy */
ReleaseWriteLock(&tdc->mflock);
goto tagain;
}
ConvertWToSLock(&tdc->mflock);
}
code = 0;
ConvertSToRLock(&tdc->mflock);
while (!code && tdc->mflags & DFFetchReq) {
afs_Trace4(afs_iclSetp, CM_TRACE_DCACHEWAIT,
ICL_TYPE_STRING, __FILE__, ICL_TYPE_INT32,
__LINE__, ICL_TYPE_POINTER, tdc,
ICL_TYPE_INT32, tdc->dflags);
/* don't need waiting flag on this one */
ReleaseReadLock(&tdc->mflock);
ReleaseReadLock(&tdc->lock);
ReleaseReadLock(&avc->lock);
code = afs_osi_SleepSig(&tdc->validPos);
ObtainReadLock(&avc->lock);
ObtainReadLock(&tdc->lock);
ObtainReadLock(&tdc->mflock);
}
ReleaseReadLock(&tdc->mflock);
if (code) {
error = code;
break;
}
}
}
/* now data may have started flowing in (if DFFetching is on). If
* data is now streaming in, then wait for some interesting stuff.
*/
code = 0;
while (!code && (tdc->dflags & DFFetching)
&& tdc->validPos <= filePos) {
/* too early: wait for DFFetching flag to vanish,
* or data to appear */
afs_Trace4(afs_iclSetp, CM_TRACE_DCACHEWAIT, ICL_TYPE_STRING,
__FILE__, ICL_TYPE_INT32, __LINE__,
ICL_TYPE_POINTER, tdc, ICL_TYPE_INT32,
tdc->dflags);
ReleaseReadLock(&tdc->lock);
ReleaseReadLock(&avc->lock);
code = afs_osi_SleepSig(&tdc->validPos);
ObtainReadLock(&avc->lock);
ObtainReadLock(&tdc->lock);
}
if (code) {
error = code;
break;
}
/* fetching flag gone, data is here, or we never tried
* (BBusy for instance) */
len = tdc->validPos - filePos;
versionOk = hsame(avc->f.m.DataVersion, tdc->f.versionNo) ? 1 : 0;
if (tdc->dflags & DFFetching) {
/* still fetching, some new data is here:
* compute length and offset */
offset = filePos - AFS_CHUNKTOBASE(tdc->f.chunk);
} else {
/* no longer fetching, verify data version (avoid new
* GetDCache call) */
if (versionOk && len > 0) {
offset = filePos - AFS_CHUNKTOBASE(tdc->f.chunk);
} else {
/* don't have current data, so get it below */
afs_Trace3(afs_iclSetp, CM_TRACE_VERSIONNO,
ICL_TYPE_INT64, ICL_HANDLE_OFFSET(filePos),
ICL_TYPE_HYPER, &avc->f.m.DataVersion,
ICL_TYPE_HYPER, &tdc->f.versionNo);
#if 0
#ifdef STRUCT_TASK_STRUCT_HAS_CRED
if (afs_protocols & VICEP_ACCESS) {
printf("afs_read: DV mismatch? %d instead of %d for %u.%u.%u\n", tdc->f.versionNo.low, avc->f.m.DataVersion.low, avc->f.fid.Fid.Volume, avc->f.fid.Fid.Vnode, avc->f.fid.Fid.Unique);
printf("afs_read: validPos %llu filePos %llu totalLength %d m.Length %llu noLock %d\n", tdc->validPos, filePos, totalLength, avc->f.m.Length, noLock);
printf("afs_read: or len too low? %lld for %u.%u.%u\n", len, avc->f.fid.Fid.Volume, avc->f.fid.Fid.Vnode, avc->f.fid.Fid.Unique);
}
#endif
#endif
ReleaseReadLock(&tdc->lock);
afs_PutDCache(tdc);
tdc = NULL;
}
}
if (!tdc) {
#ifdef STRUCT_TASK_STRUCT_HAS_CRED
if (afs_protocols & VICEP_ACCESS) { /* avoid foreground fetch */
if (!versionOk) {
printf("afs_read: avoid forground %u.%u.%u\n", avc->f.fid.Fid.Volume, avc->f.fid.Fid.Vnode, avc->f.fid.Fid.Unique);
goto try_background;
}
}
#endif
/* If we get here, it was not possible to start the
* background daemon. With flag == 1 afs_GetDCache
* does the FetchData rpc synchronously.
*/
ReleaseReadLock(&avc->lock);
tdc = afs_GetDCache(avc, filePos, treq, &offset, &len, 1);
ObtainReadLock(&avc->lock);
if (tdc)
ObtainReadLock(&tdc->lock);
}
}
if (!tdc) {
error = EIO;
break;
}
len = tdc->validPos - filePos;
afs_Trace3(afs_iclSetp, CM_TRACE_VNODEREAD, ICL_TYPE_POINTER, tdc,
ICL_TYPE_OFFSET, ICL_HANDLE_OFFSET(offset),
ICL_TYPE_OFFSET, ICL_HANDLE_OFFSET(len));
if (len > totalLength)
len = totalLength; /* will read len bytes */
if (len <= 0) { /* shouldn't get here if DFFetching is on */
afs_Trace4(afs_iclSetp, CM_TRACE_VNODEREAD2, ICL_TYPE_POINTER,
tdc, ICL_TYPE_OFFSET, ICL_HANDLE_OFFSET(tdc->validPos),
ICL_TYPE_INT32, tdc->f.chunkBytes, ICL_TYPE_INT32,
tdc->dflags);
/* read past the end of a chunk, may not be at next chunk yet, and yet
* also not at eof, so may have to supply fake zeros */
len = AFS_CHUNKTOSIZE(tdc->f.chunk) - offset; /* bytes left in chunk addr space */
if (len > totalLength)
len = totalLength; /* and still within xfr request */
tlen = avc->f.m.Length - offset; /* and still within file */
if (len > tlen)
len = tlen;
if (len > AFS_ZEROS)
len = sizeof(afs_zeros); /* and in 0 buffer */
#ifdef AFS_DARWIN80_ENV
trimlen = len;
tuiop = afsio_darwin_partialcopy(auio, trimlen);
#else
afsio_copy(auio, &tuio, tvec);
trimlen = len;
afsio_trim(&tuio, trimlen);
#endif
AFS_UIOMOVE(afs_zeros, trimlen, UIO_READ, tuiop, code);
if (code) {
error = code;
break;
}
} else {
/* get the data from the file */
tfile = (struct osi_file *)osi_UFSOpen(&tdc->f.inode);
#ifdef AFS_DARWIN80_ENV
trimlen = len;
tuiop = afsio_darwin_partialcopy(auio, trimlen);
uio_setoffset(tuiop, offset);
#else
/* mung uio structure to be right for this transfer */
afsio_copy(auio, &tuio, tvec);
trimlen = len;
afsio_trim(&tuio, trimlen);
tuio.afsio_offset = offset;
#endif
#if defined(AFS_AIX41_ENV)
AFS_GUNLOCK();
code =
VNOP_RDWR(tfile->vnode, UIO_READ, FREAD, &tuio, NULL, NULL,
NULL, afs_osi_credp);
AFS_GLOCK();
#elif defined(AFS_AIX32_ENV)
code =
VNOP_RDWR(tfile->vnode, UIO_READ, FREAD, &tuio, NULL, NULL);
/* Flush all JFS pages now for big performance gain in big file cases
* If we do something like this, must check to be sure that AFS file
* isn't mmapped... see afs_gn_map() for why.
*/
/*
if (tfile->vnode->v_gnode && tfile->vnode->v_gnode->gn_seg) {
many different ways to do similar things:
so far, the best performing one is #2, but #1 might match it if we
straighten out the confusion regarding which pages to flush. It
really does matter.
1. vm_flushp(tfile->vnode->v_gnode->gn_seg, 0, len/PAGESIZE - 1);
2. vm_releasep(tfile->vnode->v_gnode->gn_seg, offset/PAGESIZE,
(len + PAGESIZE-1)/PAGESIZE);
3. vms_inactive(tfile->vnode->v_gnode->gn_seg) Doesn't work correctly
4. vms_delete(tfile->vnode->v_gnode->gn_seg) probably also fails
tfile->vnode->v_gnode->gn_seg = NULL;
5. deletep
6. ipgrlse
7. ifreeseg
Unfortunately, this seems to cause frequent "cache corruption" episodes.
vm_releasep(tfile->vnode->v_gnode->gn_seg, offset/PAGESIZE,
(len + PAGESIZE-1)/PAGESIZE);
}
*/
#elif defined(AFS_AIX_ENV)
code =
VNOP_RDWR(tfile->vnode, UIO_READ, FREAD, (off_t) & offset,
&tuio, NULL, NULL, -1);
#elif defined(AFS_SUN5_ENV)
AFS_GUNLOCK();
#ifdef AFS_SUN510_ENV
VOP_RWLOCK(tfile->vnode, 0, NULL);
code = VOP_READ(tfile->vnode, &tuio, 0, afs_osi_credp, NULL);
VOP_RWUNLOCK(tfile->vnode, 0, NULL);
#else
VOP_RWLOCK(tfile->vnode, 0);
code = VOP_READ(tfile->vnode, &tuio, 0, afs_osi_credp);
VOP_RWUNLOCK(tfile->vnode, 0);
#endif
AFS_GLOCK();
#elif defined(AFS_SGI_ENV)
AFS_GUNLOCK();
AFS_VOP_RWLOCK(tfile->vnode, VRWLOCK_READ);
AFS_VOP_READ(tfile->vnode, &tuio, IO_ISLOCKED, afs_osi_credp,
code);
AFS_VOP_RWUNLOCK(tfile->vnode, VRWLOCK_READ);
AFS_GLOCK();
#elif defined(AFS_HPUX100_ENV)
AFS_GUNLOCK();
code = VOP_RDWR(tfile->vnode, &tuio, UIO_READ, 0, afs_osi_credp);
AFS_GLOCK();
#elif defined(AFS_LINUX20_ENV)
AFS_GUNLOCK();
code = osi_rdwr(tfile, &tuio, UIO_READ);
AFS_GLOCK();
#elif defined(AFS_DARWIN80_ENV)
AFS_GUNLOCK();
code = VNOP_READ(tfile->vnode, tuiop, 0, afs_osi_ctxtp);
AFS_GLOCK();
#elif defined(AFS_DARWIN_ENV)
AFS_GUNLOCK();
VOP_LOCK(tfile->vnode, LK_EXCLUSIVE, current_proc());
code = VOP_READ(tfile->vnode, &tuio, 0, afs_osi_credp);
VOP_UNLOCK(tfile->vnode, 0, current_proc());
AFS_GLOCK();
#elif defined(AFS_FBSD80_ENV)
AFS_GUNLOCK();
VOP_LOCK(tfile->vnode, LK_EXCLUSIVE);
code = VOP_READ(tfile->vnode, &tuio, 0, afs_osi_credp);
VOP_UNLOCK(tfile->vnode, 0);
AFS_GLOCK();
#elif defined(AFS_FBSD_ENV)
AFS_GUNLOCK();
VOP_LOCK(tfile->vnode, LK_EXCLUSIVE, curthread);
code = VOP_READ(tfile->vnode, &tuio, 0, afs_osi_credp);
VOP_UNLOCK(tfile->vnode, 0, curthread);
AFS_GLOCK();
#elif defined(AFS_NBSD_ENV)
AFS_GUNLOCK();
VOP_LOCK(tfile->vnode, LK_EXCLUSIVE);
code = VOP_READ(tfile->vnode, &tuio, 0, afs_osi_credp);
VOP_UNLOCK(tfile->vnode, 0);
AFS_GLOCK();
#elif defined(AFS_XBSD_ENV)
AFS_GUNLOCK();
VOP_LOCK(tfile->vnode, LK_EXCLUSIVE, curproc);
code = VOP_READ(tfile->vnode, &tuio, 0, afs_osi_credp);
VOP_UNLOCK(tfile->vnode, 0, curproc);
AFS_GLOCK();
#else
code = VOP_RDWR(tfile->vnode, &tuio, UIO_READ, 0, afs_osi_credp);
#endif
osi_UFSClose(tfile);
if (code) {
error = code;
break;
}
}
/* otherwise we've read some, fixup length, etc and continue with next seg */
len = len - AFS_UIO_RESID(tuiop); /* compute amount really transferred */
trimlen = len;
afsio_skip(auio, trimlen); /* update input uio structure */
totalLength -= len;
transferLength += len;
filePos += len;
if (len <= 0)
break; /* surprise eof */
#ifdef AFS_DARWIN80_ENV
if (tuiop) {
uio_free(tuiop);
tuiop = 0;
}
#endif
}
/* if we make it here with tdc non-zero, then it is the last chunk we
* dealt with, and we have to release it when we're done. We hold on
* to it in case we need to do a prefetch, obviously.
*/
if (tdc) {
ReleaseReadLock(&tdc->lock);
#if !defined(AFS_VM_RDWR_ENV)
/*
* try to queue prefetch, if needed. If DataVersion is zero there
* should not be any more: files with DV 0 never have been stored
* on the fileserver, symbolic links and directories never require
* more than a single chunk.
*/
if (!noLock && !(hiszero(avc->f.m.DataVersion))) {
if (!(tdc->mflags & DFNextStarted))
afs_PrefetchChunk(avc, tdc, acred, treq);
}
#endif
afs_PutDCache(tdc);
}
if (!noLock)
ReleaseReadLock(&avc->lock);
#ifdef AFS_DARWIN80_ENV
if (tuiop)
uio_free(tuiop);
#else
osi_FreeSmallSpace(tvec);
#endif
AFS_DISCON_UNLOCK();
error = afs_CheckCode(error, treq, 13);
afs_DestroyReq(treq);
return error;
}
int
nandfs_get_segment_info_filter(struct nandfs_device *fsdev,
struct nandfs_suinfo *nsi, uint32_t nmembs, uint64_t segment,
uint64_t *nsegs, uint32_t filter, uint32_t nfilter)
{
struct nandfs_segment_usage *su;
struct nandfs_node *su_node;
struct buf *bp;
uint64_t curr, blocknr, blockoff, i;
uint32_t flags;
int err = 0;
curr = ~(0);
lockmgr(&fsdev->nd_seg_const, LK_EXCLUSIVE, NULL);
su_node = fsdev->nd_su_node;
VOP_LOCK(NTOV(su_node), LK_SHARED);
bp = NULL;
if (nsegs != NULL)
*nsegs = 0;
for (i = 0; i < nmembs; segment++) {
if (segment == fsdev->nd_fsdata.f_nsegments)
break;
nandfs_seg_usage_blk_offset(fsdev, segment, &blocknr,
&blockoff);
if (i == 0 || curr != blocknr) {
if (bp != NULL)
brelse(bp);
err = nandfs_bread(su_node, blocknr, NOCRED,
0, &bp);
if (err) {
goto out;
}
curr = blocknr;
}
su = SU_USAGE_OFF(bp, blockoff);
flags = su->su_flags;
if (segment == fsdev->nd_seg_num ||
segment == fsdev->nd_next_seg_num)
flags |= NANDFS_SEGMENT_USAGE_ACTIVE;
if (nfilter != 0 && (flags & nfilter) != 0)
continue;
if (filter != 0 && (flags & filter) == 0)
continue;
nsi->nsi_num = segment;
nsi->nsi_lastmod = su->su_lastmod;
nsi->nsi_blocks = su->su_nblocks;
nsi->nsi_flags = flags;
nsi++;
i++;
if (nsegs != NULL)
(*nsegs)++;
}
out:
if (bp != NULL)
brelse(bp);
VOP_UNLOCK(NTOV(su_node), 0);
lockmgr(&fsdev->nd_seg_const, LK_RELEASE, NULL);
return (err);
}
/*
* Look up a vnode/nfsnode by file handle.
* Callers must check for mount points!!
* In all cases, a pointer to a
* nfsnode structure is returned.
*/
int
nfs_nget1(struct mount *mntp, nfsfh_t *fhp, int fhsize, struct nfsnode **npp,
int lkflags)
{
struct nfsnode *np;
struct vnode *vp;
struct nfsmount *nmp = VFSTONFS(mntp);
int error;
struct fh_match fhm;
fhm.fhm_fhp = fhp;
fhm.fhm_fhsize = fhsize;
loop:
rw_enter(&nmp->nm_rbtlock, RW_READER);
np = rb_tree_find_node(&nmp->nm_rbtree, &fhm);
if (np != NULL) {
vp = NFSTOV(np);
mutex_enter(vp->v_interlock);
rw_exit(&nmp->nm_rbtlock);
error = vget(vp, LK_EXCLUSIVE | lkflags);
if (error == EBUSY)
return error;
if (error)
goto loop;
*npp = np;
return(0);
}
rw_exit(&nmp->nm_rbtlock);
error = getnewvnode(VT_NFS, mntp, nfsv2_vnodeop_p, NULL, &vp);
if (error) {
*npp = 0;
return (error);
}
np = pool_get(&nfs_node_pool, PR_WAITOK);
memset(np, 0, sizeof *np);
np->n_vnode = vp;
/*
* Insert the nfsnode in the hash queue for its new file handle
*/
if (fhsize > NFS_SMALLFH) {
np->n_fhp = kmem_alloc(fhsize, KM_SLEEP);
} else
np->n_fhp = &np->n_fh;
memcpy(np->n_fhp, fhp, fhsize);
np->n_fhsize = fhsize;
np->n_accstamp = -1;
np->n_vattr = pool_get(&nfs_vattr_pool, PR_WAITOK);
rw_enter(&nmp->nm_rbtlock, RW_WRITER);
if (NULL != rb_tree_find_node(&nmp->nm_rbtree, &fhm)) {
rw_exit(&nmp->nm_rbtlock);
if (fhsize > NFS_SMALLFH) {
kmem_free(np->n_fhp, fhsize);
}
pool_put(&nfs_vattr_pool, np->n_vattr);
pool_put(&nfs_node_pool, np);
ungetnewvnode(vp);
goto loop;
}
vp->v_data = np;
genfs_node_init(vp, &nfs_genfsops);
/*
* Initalize read/write creds to useful values. VOP_OPEN will
* overwrite these.
*/
np->n_rcred = curlwp->l_cred;
kauth_cred_hold(np->n_rcred);
np->n_wcred = curlwp->l_cred;
kauth_cred_hold(np->n_wcred);
VOP_LOCK(vp, LK_EXCLUSIVE);
NFS_INVALIDATE_ATTRCACHE(np);
uvm_vnp_setsize(vp, 0);
(void)rb_tree_insert_node(&nmp->nm_rbtree, np);
rw_exit(&nmp->nm_rbtlock);
*npp = np;
return (0);
}
int
nandfs_vblock_alloc(struct nandfs_device *nandfsdev, nandfs_daddr_t *vblock)
{
struct nandfs_node *dat;
struct nandfs_mdt *mdt;
struct nandfs_alloc_request req;
struct nandfs_dat_entry *dat_entry;
uint64_t start;
uint32_t entry;
int locked, error;
dat = nandfsdev->nd_dat_node;
mdt = &nandfsdev->nd_dat_mdt;
start = nandfsdev->nd_last_cno + 1;
locked = NANDFS_VOP_ISLOCKED(NTOV(dat));
if (!locked)
VOP_LOCK(NTOV(dat), LK_EXCLUSIVE);
req.entrynum = 0;
/* Alloc vblock number */
error = nandfs_find_free_entry(mdt, dat, &req);
if (error) {
nandfs_error("%s: cannot find free vblk entry\n",
__func__);
if (!locked)
VOP_UNLOCK(NTOV(dat), 0);
return (error);
}
/* Read/create buffer */
error = nandfs_get_entry_block(mdt, dat, &req, &entry, 1);
if (error) {
nandfs_error("%s: cannot get free vblk entry\n",
__func__);
nandfs_abort_entry(&req);
if (!locked)
VOP_UNLOCK(NTOV(dat), 0);
return (error);
}
/* Fill out vblock data */
dat_entry = (struct nandfs_dat_entry *) req.bp_entry->b_data;
dat_entry[entry].de_start = start;
dat_entry[entry].de_end = UINTMAX_MAX;
dat_entry[entry].de_blocknr = 0;
/* Commit allocation */
error = nandfs_alloc_entry(mdt, &req);
if (error) {
nandfs_error("%s: cannot get free vblk entry\n",
__func__);
if (!locked)
VOP_UNLOCK(NTOV(dat), 0);
return (error);
}
/* Return allocated vblock */
*vblock = req.entrynum;
DPRINTF(DAT, ("%s: allocated vblock %#jx\n",
__func__, (uintmax_t)*vblock));
if (!locked)
VOP_UNLOCK(NTOV(dat), 0);
return (error);
}
errno_t
sc_open(thread_t *p, syscall_result_t *r, sc_open_args *arg)
{
int fd;
int flags = arg->flags;
int error = 0;
vnode_t *node;
proc_t *proc = p->thr_proc;
char fname[PATH_MAX+1];
if((error = copyinstr(fname, arg->fname, PATH_MAX)))
return error;
KASSERT(proc->p_rootdir!=NULL);
error = vfs_lookup(proc->p_curdir, &node, fname, p, LKP_NORMAL);
if(error) {
if(!(flags & O_CREAT))
return error;
vnode_t *parent;
error = vfs_lookup_parent(proc->p_curdir, &parent, fname, p);
if(error)
return error;
if((error = VOP_ACCESS(parent, W_OK, proc->p_cred))) {
vrele(parent);
return error;
}
vattr_t attr;
attr.va_mode = arg->mode & ~(proc->p_umask) & 0777;
attr.va_type = VNODE_TYPE_REG;
attr.va_uid = proc->p_cred->p_uid;
attr.va_gid = proc->p_cred->p_gid;
attr.va_size = 0;
attr.va_dev = NULL;
error = VOP_CREATE(parent, &node, _get_last_cmpt(fname), &attr);
VOP_UNLOCK(parent);
if(error)
return error;
VOP_LOCK(node);
} else {
//plik istnieje
if((flags & O_CREAT) && (flags & O_EXCL)) {
vrele(node);
return -EEXIST;
}
if((node->v_type == VNODE_TYPE_DIR) && (flags & (O_RDWR | O_WRONLY))) {
vrele(node);
return -EISDIR;
}
}
int wmode = (flags & O_RDWR) ? (W_OK | R_OK) :
(flags & O_RDONLY) ? (R_OK) : (W_OK);
if((error = VOP_ACCESS(node, wmode, proc->p_cred))) {
vrele(node);
return error;
}
if((error = VOP_OPEN(node, flags, arg->mode))) {
vrele(node);
return error;
}
if(ISSET(flags, O_RDWR | O_WRONLY) && ISSET(flags, O_TRUNC)
&& node->v_type == VNODE_TYPE_REG)
VOP_TRUNCATE(node, 0);
if((error = f_alloc(proc, node, flags, &fd))) {
vrele(node); // <- powinno to tutaj być, dopisałem bo nie było.
return error;
}
VOP_UNLOCK(node);
r->result = fd;
return 0;
}
int
afs_UFSRead(register struct vcache *avc, struct uio *auio,
struct AFS_UCRED *acred, daddr_t albn, struct buf **abpp,
int noLock)
{
afs_size_t totalLength;
afs_size_t transferLength;
afs_size_t filePos;
afs_size_t offset, len, tlen;
afs_int32 trimlen;
struct dcache *tdc = 0;
afs_int32 error;
#ifdef AFS_DARWIN80_ENV
uio_t tuiop=NULL;
#else
struct uio tuio;
struct uio *tuiop = &tuio;
struct iovec *tvec;
#endif
struct osi_file *tfile;
afs_int32 code;
int trybusy = 1;
struct vrequest treq;
AFS_STATCNT(afs_UFSRead);
if (avc && avc->vc_error)
return EIO;
AFS_DISCON_LOCK();
/* check that we have the latest status info in the vnode cache */
if ((code = afs_InitReq(&treq, acred)))
return code;
if (!noLock) {
if (!avc)
osi_Panic("null avc in afs_UFSRead");
else {
code = afs_VerifyVCache(avc, &treq);
if (code) {
code = afs_CheckCode(code, &treq, 11); /* failed to get it */
AFS_DISCON_UNLOCK();
return code;
}
}
}
#ifndef AFS_VM_RDWR_ENV
if (AFS_NFSXLATORREQ(acred)) {
if (!afs_AccessOK
(avc, PRSFS_READ, &treq,
CHECK_MODE_BITS | CMB_ALLOW_EXEC_AS_READ)) {
AFS_DISCON_UNLOCK();
return afs_CheckCode(EACCES, &treq, 12);
}
}
#endif
#ifndef AFS_DARWIN80_ENV
tvec = (struct iovec *)osi_AllocSmallSpace(sizeof(struct iovec));
#endif
totalLength = AFS_UIO_RESID(auio);
filePos = AFS_UIO_OFFSET(auio);
afs_Trace4(afs_iclSetp, CM_TRACE_READ, ICL_TYPE_POINTER, avc,
ICL_TYPE_OFFSET, ICL_HANDLE_OFFSET(filePos), ICL_TYPE_INT32,
totalLength, ICL_TYPE_OFFSET,
ICL_HANDLE_OFFSET(avc->f.m.Length));
error = 0;
transferLength = 0;
if (!noLock)
ObtainReadLock(&avc->lock);
#if defined(AFS_TEXT_ENV) && !defined(AFS_VM_RDWR_ENV)
if (avc->flushDV.high == AFS_MAXDV && avc->flushDV.low == AFS_MAXDV) {
hset(avc->flushDV, avc->f.m.DataVersion);
}
#endif
if (filePos >= avc->f.m.Length) {
if (len > AFS_ZEROS)
len = sizeof(afs_zeros); /* and in 0 buffer */
len = 0;
#ifdef AFS_DARWIN80_ENV
trimlen = len;
tuiop = afsio_darwin_partialcopy(auio, trimlen);
#else
afsio_copy(auio, &tuio, tvec);
trimlen = len;
afsio_trim(&tuio, trimlen);
#endif
AFS_UIOMOVE(afs_zeros, trimlen, UIO_READ, tuiop, code);
}
while (avc->f.m.Length > 0 && totalLength > 0) {
/* read all of the cached info */
if (filePos >= avc->f.m.Length)
break; /* all done */
if (noLock) {
if (tdc) {
ReleaseReadLock(&tdc->lock);
afs_PutDCache(tdc);
}
tdc = afs_FindDCache(avc, filePos);
if (tdc) {
ObtainReadLock(&tdc->lock);
offset = filePos - AFS_CHUNKTOBASE(tdc->f.chunk);
len = tdc->validPos - filePos;
}
} else {
/* a tricky question: does the presence of the DFFetching flag
* mean that we're fetching the latest version of the file? No.
* The server could update the file as soon as the fetch responsible
* for the setting of the DFFetching flag completes.
*
* However, the presence of the DFFetching flag (visible under
* a dcache read lock since it is set and cleared only under a
* dcache write lock) means that we're fetching as good a version
* as was known to this client at the time of the last call to
* afs_VerifyVCache, since the latter updates the stat cache's
* m.DataVersion field under a vcache write lock, and from the
* time that the DFFetching flag goes on in afs_GetDCache (before
* the fetch starts), to the time it goes off (after the fetch
* completes), afs_GetDCache keeps at least a read lock on the
* vcache entry.
*
* This means that if the DFFetching flag is set, we can use that
* data for any reads that must come from the current version of
* the file (current == m.DataVersion).
*
* Another way of looking at this same point is this: if we're
* fetching some data and then try do an afs_VerifyVCache, the
* VerifyVCache operation will not complete until after the
* DFFetching flag is turned off and the dcache entry's f.versionNo
* field is updated.
*
* Note, by the way, that if DFFetching is set,
* m.DataVersion > f.versionNo (the latter is not updated until
* after the fetch completes).
*/
if (tdc) {
ReleaseReadLock(&tdc->lock);
afs_PutDCache(tdc); /* before reusing tdc */
}
tdc = afs_GetDCache(avc, filePos, &treq, &offset, &len, 2);
#ifdef AFS_DISCON_ENV
if (!tdc) {
printf("Network down in afs_read");
error = ENETDOWN;
break;
}
#endif /* AFS_DISCON_ENV */
ObtainReadLock(&tdc->lock);
/* now, first try to start transfer, if we'll need the data. If
* data already coming, we don't need to do this, obviously. Type
* 2 requests never return a null dcache entry, btw. */
if (!(tdc->dflags & DFFetching)
&& !hsame(avc->f.m.DataVersion, tdc->f.versionNo)) {
/* have cache entry, it is not coming in now, and we'll need new data */
tagain:
if (trybusy && !afs_BBusy()) {
struct brequest *bp;
/* daemon is not busy */
ObtainSharedLock(&tdc->mflock, 667);
if (!(tdc->mflags & DFFetchReq)) {
UpgradeSToWLock(&tdc->mflock, 668);
tdc->mflags |= DFFetchReq;
bp = afs_BQueue(BOP_FETCH, avc, B_DONTWAIT, 0, acred,
(afs_size_t) filePos, (afs_size_t) 0,
tdc);
if (!bp) {
/* Bkg table full; retry deadlocks */
tdc->mflags &= ~DFFetchReq;
trybusy = 0; /* Avoid bkg daemon since they're too busy */
ReleaseWriteLock(&tdc->mflock);
goto tagain;
}
ConvertWToSLock(&tdc->mflock);
}
code = 0;
ConvertSToRLock(&tdc->mflock);
while (!code && tdc->mflags & DFFetchReq) {
afs_Trace4(afs_iclSetp, CM_TRACE_DCACHEWAIT,
ICL_TYPE_STRING, __FILE__, ICL_TYPE_INT32,
__LINE__, ICL_TYPE_POINTER, tdc,
ICL_TYPE_INT32, tdc->dflags);
/* don't need waiting flag on this one */
ReleaseReadLock(&tdc->mflock);
ReleaseReadLock(&tdc->lock);
ReleaseReadLock(&avc->lock);
code = afs_osi_SleepSig(&tdc->validPos);
ObtainReadLock(&avc->lock);
ObtainReadLock(&tdc->lock);
ObtainReadLock(&tdc->mflock);
}
ReleaseReadLock(&tdc->mflock);
if (code) {
error = code;
break;
}
}
}
/* now data may have started flowing in (if DFFetching is on). If
* data is now streaming in, then wait for some interesting stuff.
*/
code = 0;
while (!code && (tdc->dflags & DFFetching)
&& tdc->validPos <= filePos) {
/* too early: wait for DFFetching flag to vanish,
* or data to appear */
afs_Trace4(afs_iclSetp, CM_TRACE_DCACHEWAIT, ICL_TYPE_STRING,
__FILE__, ICL_TYPE_INT32, __LINE__,
ICL_TYPE_POINTER, tdc, ICL_TYPE_INT32,
tdc->dflags);
ReleaseReadLock(&tdc->lock);
ReleaseReadLock(&avc->lock);
code = afs_osi_SleepSig(&tdc->validPos);
ObtainReadLock(&avc->lock);
ObtainReadLock(&tdc->lock);
}
if (code) {
error = code;
break;
}
/* fetching flag gone, data is here, or we never tried
* (BBusy for instance) */
if (tdc->dflags & DFFetching) {
/* still fetching, some new data is here:
* compute length and offset */
offset = filePos - AFS_CHUNKTOBASE(tdc->f.chunk);
len = tdc->validPos - filePos;
} else {
/* no longer fetching, verify data version (avoid new
* GetDCache call) */
if (hsame(avc->f.m.DataVersion, tdc->f.versionNo)
&& ((len = tdc->validPos - filePos) > 0)) {
offset = filePos - AFS_CHUNKTOBASE(tdc->f.chunk);
} else {
/* don't have current data, so get it below */
afs_Trace3(afs_iclSetp, CM_TRACE_VERSIONNO,
ICL_TYPE_INT64, ICL_HANDLE_OFFSET(filePos),
ICL_TYPE_HYPER, &avc->f.m.DataVersion,
ICL_TYPE_HYPER, &tdc->f.versionNo);
ReleaseReadLock(&tdc->lock);
afs_PutDCache(tdc);
tdc = NULL;
}
}
if (!tdc) {
/* If we get, it was not possible to start the
* background daemon. With flag == 1 afs_GetDCache
* does the FetchData rpc synchronously.
*/
ReleaseReadLock(&avc->lock);
tdc = afs_GetDCache(avc, filePos, &treq, &offset, &len, 1);
ObtainReadLock(&avc->lock);
if (tdc)
ObtainReadLock(&tdc->lock);
}
}
if (!tdc) {
error = EIO;
break;
}
len = tdc->validPos - filePos;
afs_Trace3(afs_iclSetp, CM_TRACE_VNODEREAD, ICL_TYPE_POINTER, tdc,
ICL_TYPE_OFFSET, ICL_HANDLE_OFFSET(offset),
ICL_TYPE_OFFSET, ICL_HANDLE_OFFSET(len));
if (len > totalLength)
len = totalLength; /* will read len bytes */
if (len <= 0) { /* shouldn't get here if DFFetching is on */
afs_Trace4(afs_iclSetp, CM_TRACE_VNODEREAD2, ICL_TYPE_POINTER,
tdc, ICL_TYPE_OFFSET, ICL_HANDLE_OFFSET(tdc->validPos),
ICL_TYPE_INT32, tdc->f.chunkBytes, ICL_TYPE_INT32,
tdc->dflags);
/* read past the end of a chunk, may not be at next chunk yet, and yet
* also not at eof, so may have to supply fake zeros */
len = AFS_CHUNKTOSIZE(tdc->f.chunk) - offset; /* bytes left in chunk addr space */
if (len > totalLength)
len = totalLength; /* and still within xfr request */
tlen = avc->f.m.Length - offset; /* and still within file */
if (len > tlen)
len = tlen;
if (len > AFS_ZEROS)
len = sizeof(afs_zeros); /* and in 0 buffer */
#ifdef AFS_DARWIN80_ENV
trimlen = len;
tuiop = afsio_darwin_partialcopy(auio, trimlen);
#else
afsio_copy(auio, &tuio, tvec);
trimlen = len;
afsio_trim(&tuio, trimlen);
#endif
AFS_UIOMOVE(afs_zeros, trimlen, UIO_READ, tuiop, code);
if (code) {
error = code;
break;
}
} else {
/* get the data from the file */
#ifdef IHINT
if (tfile = tdc->ihint) {
if (tdc->f.inode != tfile->inum) {
afs_warn("afs_UFSRead: %x hint mismatch tdc %d inum %d\n",
tdc, tdc->f.inode, tfile->inum);
osi_UFSClose(tfile);
tdc->ihint = tfile = 0;
nihints--;
}
}
if (tfile != 0) {
usedihint++;
} else
#endif /* IHINT */
#if defined(LINUX_USE_FH)
tfile = (struct osi_file *)osi_UFSOpen_fh(&tdc->f.fh, tdc->f.fh_type);
#else
tfile = (struct osi_file *)osi_UFSOpen(tdc->f.inode);
#endif
#ifdef AFS_DARWIN80_ENV
trimlen = len;
tuiop = afsio_darwin_partialcopy(auio, trimlen);
uio_setoffset(tuiop, offset);
#else
/* mung uio structure to be right for this transfer */
afsio_copy(auio, &tuio, tvec);
trimlen = len;
afsio_trim(&tuio, trimlen);
tuio.afsio_offset = offset;
#endif
#if defined(AFS_AIX41_ENV)
AFS_GUNLOCK();
code =
VNOP_RDWR(tfile->vnode, UIO_READ, FREAD, &tuio, NULL, NULL,
NULL, afs_osi_credp);
AFS_GLOCK();
#elif defined(AFS_AIX32_ENV)
code =
VNOP_RDWR(tfile->vnode, UIO_READ, FREAD, &tuio, NULL, NULL);
/* Flush all JFS pages now for big performance gain in big file cases
* If we do something like this, must check to be sure that AFS file
* isn't mmapped... see afs_gn_map() for why.
*/
/*
if (tfile->vnode->v_gnode && tfile->vnode->v_gnode->gn_seg) {
many different ways to do similar things:
so far, the best performing one is #2, but #1 might match it if we
straighten out the confusion regarding which pages to flush. It
really does matter.
1. vm_flushp(tfile->vnode->v_gnode->gn_seg, 0, len/PAGESIZE - 1);
2. vm_releasep(tfile->vnode->v_gnode->gn_seg, offset/PAGESIZE,
(len + PAGESIZE-1)/PAGESIZE);
3. vms_inactive(tfile->vnode->v_gnode->gn_seg) Doesn't work correctly
4. vms_delete(tfile->vnode->v_gnode->gn_seg) probably also fails
tfile->vnode->v_gnode->gn_seg = NULL;
5. deletep
6. ipgrlse
7. ifreeseg
Unfortunately, this seems to cause frequent "cache corruption" episodes.
vm_releasep(tfile->vnode->v_gnode->gn_seg, offset/PAGESIZE,
(len + PAGESIZE-1)/PAGESIZE);
}
*/
#elif defined(AFS_AIX_ENV)
code =
VNOP_RDWR(tfile->vnode, UIO_READ, FREAD, (off_t) & offset,
&tuio, NULL, NULL, -1);
#elif defined(AFS_SUN5_ENV)
AFS_GUNLOCK();
#ifdef AFS_SUN510_ENV
{
caller_context_t ct;
VOP_RWLOCK(tfile->vnode, 0, &ct);
code = VOP_READ(tfile->vnode, &tuio, 0, afs_osi_credp, &ct);
VOP_RWUNLOCK(tfile->vnode, 0, &ct);
}
#else
VOP_RWLOCK(tfile->vnode, 0);
code = VOP_READ(tfile->vnode, &tuio, 0, afs_osi_credp);
VOP_RWUNLOCK(tfile->vnode, 0);
#endif
AFS_GLOCK();
#elif defined(AFS_SGI_ENV)
AFS_GUNLOCK();
AFS_VOP_RWLOCK(tfile->vnode, VRWLOCK_READ);
AFS_VOP_READ(tfile->vnode, &tuio, IO_ISLOCKED, afs_osi_credp,
code);
AFS_VOP_RWUNLOCK(tfile->vnode, VRWLOCK_READ);
AFS_GLOCK();
#elif defined(AFS_OSF_ENV)
tuio.uio_rw = UIO_READ;
AFS_GUNLOCK();
VOP_READ(tfile->vnode, &tuio, 0, afs_osi_credp, code);
AFS_GLOCK();
#elif defined(AFS_HPUX100_ENV)
AFS_GUNLOCK();
code = VOP_RDWR(tfile->vnode, &tuio, UIO_READ, 0, afs_osi_credp);
AFS_GLOCK();
#elif defined(AFS_LINUX20_ENV)
AFS_GUNLOCK();
code = osi_rdwr(tfile, &tuio, UIO_READ);
AFS_GLOCK();
#elif defined(AFS_DARWIN80_ENV)
AFS_GUNLOCK();
code = VNOP_READ(tfile->vnode, tuiop, 0, afs_osi_ctxtp);
AFS_GLOCK();
#elif defined(AFS_DARWIN_ENV)
AFS_GUNLOCK();
VOP_LOCK(tfile->vnode, LK_EXCLUSIVE, current_proc());
code = VOP_READ(tfile->vnode, &tuio, 0, afs_osi_credp);
VOP_UNLOCK(tfile->vnode, 0, current_proc());
AFS_GLOCK();
#elif defined(AFS_FBSD80_ENV)
AFS_GUNLOCK();
VOP_LOCK(tfile->vnode, LK_EXCLUSIVE);
code = VOP_READ(tfile->vnode, &tuio, 0, afs_osi_credp);
VOP_UNLOCK(tfile->vnode, 0);
AFS_GLOCK();
#elif defined(AFS_FBSD50_ENV)
AFS_GUNLOCK();
VOP_LOCK(tfile->vnode, LK_EXCLUSIVE, curthread);
code = VOP_READ(tfile->vnode, &tuio, 0, afs_osi_credp);
VOP_UNLOCK(tfile->vnode, 0, curthread);
AFS_GLOCK();
#elif defined(AFS_XBSD_ENV)
AFS_GUNLOCK();
VOP_LOCK(tfile->vnode, LK_EXCLUSIVE, curproc);
code = VOP_READ(tfile->vnode, &tuio, 0, afs_osi_credp);
VOP_UNLOCK(tfile->vnode, 0, curproc);
AFS_GLOCK();
#else
code = VOP_RDWR(tfile->vnode, &tuio, UIO_READ, 0, afs_osi_credp);
#endif
#ifdef IHINT
if (!tdc->ihint && nihints < maxIHint) {
tdc->ihint = tfile;
nihints++;
} else
#endif /* IHINT */
osi_UFSClose(tfile);
if (code) {
error = code;
break;
}
}
/* otherwise we've read some, fixup length, etc and continue with next seg */
len = len - AFS_UIO_RESID(tuiop); /* compute amount really transferred */
trimlen = len;
afsio_skip(auio, trimlen); /* update input uio structure */
totalLength -= len;
transferLength += len;
filePos += len;
if (len <= 0)
break; /* surprise eof */
#ifdef AFS_DARWIN80_ENV
if (tuiop) {
uio_free(tuiop);
tuiop = 0;
}
#endif
}
/* if we make it here with tdc non-zero, then it is the last chunk we
* dealt with, and we have to release it when we're done. We hold on
* to it in case we need to do a prefetch, obviously.
*/
if (tdc) {
ReleaseReadLock(&tdc->lock);
#if !defined(AFS_VM_RDWR_ENV)
/* try to queue prefetch, if needed */
if (!noLock) {
if (!(tdc->mflags & DFNextStarted))
afs_PrefetchChunk(avc, tdc, acred, &treq);
}
#endif
afs_PutDCache(tdc);
}
if (!noLock)
ReleaseReadLock(&avc->lock);
#ifdef AFS_DARWIN80_ENV
if (tuiop)
uio_free(tuiop);
#else
osi_FreeSmallSpace(tvec);
#endif
AFS_DISCON_UNLOCK();
error = afs_CheckCode(error, &treq, 13);
return error;
}
static int
nandfs_cleaner_clean_segments(struct nandfs_device *nffsdev,
struct nandfs_vinfo *vinfo, uint32_t nvinfo,
struct nandfs_period *pd, uint32_t npd,
struct nandfs_bdesc *bdesc, uint32_t nbdesc,
uint64_t *segments, uint32_t nsegs)
{
struct nandfs_node *gc;
struct buf *bp;
uint32_t i;
int error = 0;
gc = nffsdev->nd_gc_node;
DPRINTF(CLEAN, ("%s: enter\n", __func__));
VOP_LOCK(NTOV(gc), LK_EXCLUSIVE);
for (i = 0; i < nvinfo; i++) {
if (!vinfo[i].nvi_alive)
continue;
DPRINTF(CLEAN, ("%s: read vblknr:%#jx blk:%#jx\n",
__func__, (uintmax_t)vinfo[i].nvi_vblocknr,
(uintmax_t)vinfo[i].nvi_blocknr));
error = nandfs_bread(nffsdev->nd_gc_node, vinfo[i].nvi_blocknr,
NULL, 0, &bp);
if (error) {
nandfs_error("%s:%d", __FILE__, __LINE__);
VOP_UNLOCK(NTOV(gc), 0);
goto out;
}
nandfs_vblk_set(bp, vinfo[i].nvi_vblocknr);
nandfs_buf_set(bp, NANDFS_VBLK_ASSIGNED);
nandfs_dirty_buf(bp, 1);
}
VOP_UNLOCK(NTOV(gc), 0);
/* Delete checkpoints */
for (i = 0; i < npd; i++) {
DPRINTF(CLEAN, ("delete checkpoint: %jx\n",
(uintmax_t)pd[i].p_start));
error = nandfs_delete_cp(nffsdev->nd_cp_node, pd[i].p_start,
pd[i].p_end);
if (error) {
nandfs_error("%s:%d", __FILE__, __LINE__);
goto out;
}
}
/* Update vblocks */
for (i = 0; i < nvinfo; i++) {
if (vinfo[i].nvi_alive)
continue;
DPRINTF(CLEAN, ("freeing vblknr: %jx\n", vinfo[i].nvi_vblocknr));
error = nandfs_vblock_free(nffsdev, vinfo[i].nvi_vblocknr);
if (error) {
nandfs_error("%s:%d", __FILE__, __LINE__);
goto out;
}
}
error = nandfs_process_bdesc(nffsdev, bdesc, nbdesc);
if (error) {
nandfs_error("%s:%d", __FILE__, __LINE__);
goto out;
}
/* Add segments to clean */
if (nffsdev->nd_free_count) {
nffsdev->nd_free_base = realloc(nffsdev->nd_free_base,
(nffsdev->nd_free_count + nsegs) * sizeof(uint64_t),
M_NANDFSTEMP, M_WAITOK | M_ZERO);
memcpy(&nffsdev->nd_free_base[nffsdev->nd_free_count], segments,
nsegs * sizeof(uint64_t));
nffsdev->nd_free_count += nsegs;
} else {
nffsdev->nd_free_base = malloc(nsegs * sizeof(uint64_t),
M_NANDFSTEMP, M_WAITOK|M_ZERO);
memcpy(nffsdev->nd_free_base, segments,
nsegs * sizeof(uint64_t));
nffsdev->nd_free_count = nsegs;
}
out:
DPRINTF(CLEAN, ("%s: exit error %d\n", __func__, error));
return (error);
}
